Declining R&D efficiency: Evidence from Japan

Introduction

Ordinary Japanese people, who believed their country was one of the most technologically advanced, expected Japan would quickly develop a vaccine for COVID-19. The failure to do so shocked Japanese society, raising questions about research and technological progress in the country. Building on earlier research, this essay examines two key measures that show research and development (R&D) efficiency in Japan did indeed decline in the 2010s from the 2000s.

The recent weak contribution of R&D activities to productivity growth is a major issue in the analysis of secular economic stagnation in Japan. Although Japan has maintained a ratio of R&D to gross domestic product (GDP) of 3% for some time, R&D efficiency growth appears to have slowed. Bloom et al. (2020) point to one possible solution to the puzzle. They argued that R&D efficiency, measured by economic productivity growth divided by the number of researchers, has declined in the United States. Following their work, Miyagawa and Ishikawa (2019) found that the efficiency of R&D in Japanese manufacturing and information services has also declined.

Using more recent data, this essay examines two measures of R&D efficiency. The first is derived from a simple production function in which productivity depends on the stock of R&D. The second, developed by Bloom et al. (2020), is expressed as economic productivity growth divided by labour input allocated to R&D.

Both measures show that R&D efficiency in Japan in the 2010s declined compared to the 2000s. These results suggest the Japanese government should consider further supporting investment in human resources and organisational change, both of which are complementary to R&D.

Secular stagnation and the decline of R&D efficiency

The secular stagnation in economic productivity growth in the United States after the global economic crisis of 2008 is a topic of active discussion. An optimistic view, expressed by Brynjolfsson and McAfee (2014) and Aghion et al. (2019), is that slow labour productivity growth results from a mismeasurement of quality in new technology-generated services. A pessimistic view, articulated by Gordon (2016), is that acceleration of economic productivity growth due to the information and communication technology (ICT) revolution has ended. In another interpretation, Bloom et al. (2020) argue that a decline of R&D efficiency plays a role.

Recognising the importance of R&D for economic productivity, Japan has kept the ratio of R&D to GDP at around 3% for over 20 years. Nevertheless, the Japanese economy has stagnated for a long time and has not returned to the growth rates seen in the 1980s. Miyagawa and Ishikawa (2019) measured R&D efficiency at the industry level for the 20 years from 1995 to 2015 using the Japan Industrial Productivity database (JIP database)1 and the EUKLEMS database, following the approach of Bloom et al. (2020). They found that R&D efficiency in Japan declined over this period.

Two approaches to measuring R&D efficiency

This essay considers two approaches for measuring R&D efficiency. The first, developed by Griliches (1979), recognises that accumulation of R&D expenditures constitutes a stock of knowledge that contributes to economic productivity. Assuming a standard production function, R&D efficiency in this approach is expressed as follows:

Economic productivity growth (total factor productivity growth) = R&D efficiency (marginal efficiency of knowledge stock) x R&D intensity (R&D expenditures/GDP) (1)

From equation 1, using time-series data for total factor productivity (TFP) growth and R&D intensity, changes in R&D efficiency can be examined.

The second approach was developed by Bloom et al. (2020). In the context of endogenous growth theory, economic productivity growth here depends on the number of researchers. However, instead of an actual headcount, Bloom et al. (2020) use a measure they term “effective R&D”. They obtain effective R&D by dividing R&D expenditure by an appropriate wage rate for researchers. They do this as a way to address the difficulty of measuring the total number of researchers. Accordingly, from equation 2, Bloom et al. (2020) derive R&D efficiency as follows:

R&D efficiency x effective R&D = TFP growth rate (2)

R&D efficiency in manufacturing

Table 1 shows data on changes in variables associated with measured R&D efficiency in Japanese manufacturing in the late 1990s, 2000s and 2010s. It indicates that TFP growth declined across the three periods, while R&D intensity and effective R&D increased (as noted, effective R&D is measured by dividing R&D expenditure by labour compensation per hour). When these data are included in equations (1) and (2), above, the result suggests a falling trend in R&D efficiency.

The analysis also examined the hypothesis of declining R&D efficiency using cross-sectional data. The manufacturing sector in the JIP 2021 database consists of 54 industries, which the analysis divided into two periods: from 2000-10 and 2010-18. TFP growth and effective R&D by industry were measured and plotted in Figures 1 and 2. The slope of the tangent indicates changes in R&D efficiency. The slope in the 2010s (Figure 2) is flatter than that in the 2000s (Figure 1), again suggesting declining R&D efficiency in Japan.

Figure 1. The relationship between annual average TFP growth and effective R&D by industry, 2000-18

Source: JIP database (2021), www.rieti.go.jp/en/database/JIP2021/index.html.

As a main limitation of this approach to R&D efficiency, TFP growth is affected not only by R&D activities but also by several other drivers such as human and organisational capital. Therefore, the analysis includes the number of patents as an outcome of R&D activity because this is a closer proxy of R&D than TFP growth (Hall et al., 2005).

Accordingly, the ratio of new patent applications to the total number of patents are divided into two periods: 1996-2005 and 2006-15. Effective R&D by industry is also measured from the JIP database. R&D efficiency in each period was obtained by dividing the average number of patents in each period by the average value of effective R&D in the corresponding period. This measure showed that R&D efficiency in manufacturing in the second period was 56% of that in the first period.

R&D efficiency in information services

R&D in manufacturing accounts for over 70% of all R&D spending in Japan. However, R&D expenditure in information services is the largest in the service sector overall (excluding the research and education industries). In addition, in information services, software investment has a similar role to R&D investment.

According to the JIP database, average annual TFP growth was negative in Japan’s information services industry from 2000 to 2017. As effective R&D has increased since 1995, there was negative R&D efficiency growth in information services. One possible reason for this is the slow growth of the information services market in Japan. In the United Kingdom and the United States, TFP growth rates in information services became positive in the 2000s, after negative rates in the late 1990s. In this case, companies in information services invested aggressively in R&D and software early in the ICT revolution. As the technology-productivity J-curve developed by Brynjolfsson et al. (2021) suggests, this investment in new technology in the late 1990s likely contributed to high productivity growth in the 2000s.

Conclusion

Using several measures, this essay shows that R&D efficiency in Japanese manufacturing, and in the information services industry, has declined. These findings are consistent with Bloom et al. (2020), who pointed to a decline in R&D efficiency in the United States. Japan has spent around 3% of GDP on R&D for many years. However, these results imply the scale of investment is not enough to achieve required improvements in economic productivity.